hpc user environment 1...hpc user environment 1 yuwu chen hpc user services lsu hpc loni...

HPC User Environment 1

Yuwu ChenHPC User Services

LSU HPC [email protected]

Louisiana State UniversityBaton Rouge

February 7, 2018

Outline Things to be covered in the training

– Basic HPC concepts• Computational speed evaluation• Top 500 list

– Available HPC resources• LONI & LSU HPC• Account and allocations

HPC User Environment 1 Spring 2018 2

– HPC software environment• General cluster architecture

• How to access HPC & LONI clusters• How to check file systems, allocation balance• How to transfer files between cluster and local PC• The software management tool softenv and modules

– Job submission and management basics

What is HPC

High Performance Computing (HPC) is computation at the cuttingedge of modern technology, often done on a supercomputer

A supercomputer is in the class of machines that rank among thefastest in the world– Rule of thumb: a supercomputer could be defined to be at least 100

times as powerful as a PC


How do we evaluate the performance of HPC?600 mph 60 mph

Core, CPU, Processor? A core is usually the basic computation unit. A CPU may have one or more cores to perform tasks at a given time.


In this training, CPU = processor, which has multiple cores. The term“CPU” will not be used.

Case study – core and processor


How many cores does this computer have?

4 cores/processor * 4 processors = 16 total cores

Measure HPC performance-FLOPS

Performance is measured in Floating Point Operations PerSecond (FLOPS or flop/s)

= × ×– Most processors today can do 4 FLOPs per clock cycle.

Therefore a single-core 2.5-GHz processor has atheoretical performance of 10 billion FLOPs = 10GFLOPs

– Dual core, quad core? (i3, i5, i7)


Intel i9-7980XE was released in September 2017– 18 cores– “the first teraflop desktop PC processor ever”– $1999

18 * 4.4 GHz (when Turbo Boost) *16 = 1267 GFLOPS= 1. 267 TFLOPS

Supercomputing on a Cell Phone?

Hex-core processors are coming to your phone– Nvidia, TI, QualComm…– Processing power in the neighborhood of 50 GigaFLOPS– Would make the top 500 supercomputer list 20 years

ago– What is your phone’s FLOPS?


• Compare to ENIAC (500FLOPS)

• 57.4 GFLOPS

• Compare to top 500 in 1993 #1(59.7 GFLOPS), #500 (0.42GFLOPS)

• iPhone X A11 Bionic:• 2×Monsoon + 4×Mistral (2.39

GHz)

The Top 500 List

The TOP500 project provides a list of 500 fastest super-computers in the world ranked by their LINPACK performance.

Semi-annually published (in the public domain) As of November 2017, China’s Sunway TaihuLight supercomputer

is the fastest in the world.o Cores: 10,649,600o Peak Performance: 125,436TFlop/s

As of November 2017o LSU SuperMIC ranks 491 925TFlop/so LONI QB2 ranks 152 1474TFlop/s


Intel i9-7980XE: 1.267 TFlop/s

Supercomputer Cluster Racks


Inside A Cluster Rack

HPC User Environment 1 Spring 2018

Rack

Interconnect:Infiniband

Switch

ComputeNode

10

Inside A Compute Node


NetworkCard

ProcessorMemory

Storage

Accelerator(GPU/Xeon Phi)

11


Available Computing Resources

University wide-HPC@LSU State wide-LONI Nation wide-XSEDE

12

University Level: HPC@LSU University Level: LSU HPC resources available to LSU Faculty and

their affiliates. LONI and LSU HPC administered and supported by HPC@LSU


Available LSU HPC ResourcesPhilip

Hostname philip.hpc.lsu.edu

PeakPerformance/TFlops 3.469

Compute nodes 37

Processor/node 2 Quad-Core

Processor Speed 2.93GHz

Processor Type Intel Xeon 64bit

Nodes withAccelerators 2

Accelerator Type 3 nVidia M2070

OS RHEL v5Vendor Dell

Memory per node 24/48/96 GB

Detailed Cluster Description

User GuideAvailable Software


SuperMike II

Hostname mike.hpc.lsu.edu

PeakPerformance/TFlops 146

Compute nodes 440

Processor/node 2 Octa-core




Accelerator Type 2 nVidia M2090

OS RHEL v6Vendor Dell

Memory per node 32/64/256 GB



Ref: http://www.hpc.lsu.edu/resources/hpc/index.php#lsuhpc

SuperMIC

Hostname smic.hpc.lsu.edu

PeakPerformance/TFlops 1000

Compute nodes 360

Processor/node 2 Deca-core




Accelerator Type Xeon Phi 7120P

OS RHEL v6Vendor

Memory per node 64 GB



14

State Level - Louisiana OpticalNetwork Infrastructure (LONI)

A state-of-the-art fiber optic network that runs throughout Louisianaand connects Louisiana and Mississippi research universities.

$40M Optical Network, 10Gb Ethernet over fiber optics.


LONI-Louisiana Optical Network Infrastructure

LONI connects supercomputers at Louisiana's universities:– Louisiana State University– Louisiana Tech University– LSU Health Sciences Center in New Orleans– LSU Health Sciences Center in Shreveport– Southern University– Tulane University– University of Louisiana at Lafayette– University of New Orleans– Grambling State University– Southeastern Louisiana University


Available LONI Resources


Ref: http://www.hpc.lsu.edu/resources/hpc/index.php#loni

QB2

Hostname qb2.loni.org

Peak Performance/TFlops 1,500

Compute nodes 504

Processor/node 2 10-Core


Processor Type Intel Ivy Bridge-EP Xeon 64bit

Nodes with Accelerators 480

Accelerator Type NVIDIA Tesla K20x

OS RHEL v6

Vendor Dell

Memory per node 64 GB

Location Information Systems Building,Baton Rouge


User Guide

Available Software

17

Summary of Clusters for LSU and LONI

Name Performance(TFLOPS) Location Vendor Architecture

LONI QB2 1474 ISB Dell Linux x86_64

LSUPhilip 3.5 LSU Dell Linux x86_64

SuperMIC 925 LSU Dell Linux x86_64SuperMike 212 (CPU+GPU) LSU Dell Linux x86_64


ISB: Information Services Building (Downtown Baton Rouge)

National Level National Level: Extreme Science and Engineering Discovery

Environment (XSEDE)– 5 year, $121M project supported by NSF– Supports 16 supercomputers and high-end visualization and data

analysis resources across the country.– https://www.xsede.org/– LSU SuperMIC is one of the XSEDE clusters.


Account Eligibility-LSU HPC All faculty and research staff at Louisiana State University, as well as

students pursuing sponsored research activities at LSU, are eligiblefor a LSU HPC account.


ACCOUNT SPONSOR: The person who is responsible for youractivities on the cluster.

User accounts on LSU HPC require a valid LSU email address, and LSUaffiliated sponsor. Only faculty members or research staff from LSU BatonRouge campus can sponsor accounts.

LSU students, postdoctoral researchers, or research associates may usetheir advisor as their sponsor.

For prospective LSU HPC Users from outside LSU, you are required tohave a faculty member at LSU as your Collaborator to sponsor you a LSUHPC account.

Account Eligibility-LONI All faculty and research staff at a LONI Member Institution, as well as

students pursuing sponsored research activities at these facilities, areeligible for a LONI account.

Requests for accounts by research associates not affiliated with a LONIMember Institution will be handled on a case by case basis.

For prospective LONI Users from a non-LONI Member Institution, you arerequired to have a faculty member in one of LONI Member Institutions asyour Collaborator to sponsor you a LONI account.


Account Eligibility Test I can be granted a LSU HPC or LONI account if:

a) I am using HPC resource for my research, sponsored by my PIb) I am attending HPC training sessions, sponsored by the HPC staffc) I am taking a class that requires using HPC resource, sponsored by the

course instructord) a and be) a and cf) All of the above


LONI & LSU HPC Accounts

LSU HPC and LONI systems are two distinct computational resourcesadministered by HPC@LSU.

Having an account on one does not grant the user access to the other.

Contact user services– Email Help Ticket: [email protected]– Telephone Help Desk: +1 (225) 578-0900


How Do I Get a LONI Account?

• Visit https://allocations.loni.org/login_request.php• Enter your INSTITUTIONAL Email Address.• Check your email and click on the link provided (link is active for

24hrs only)• Fill the form provided• For LONI CONTACT/COLLABORATOR field enter the name of your

research advisor/supervisor or course instructor who must be a FullTime Faculty member at a LONI member institution. Never enter HPCstaff’s name.

• Click Submit button• Your account will be activated once we have verified your credentials.


How Do I Get a LSU HPC Account?

• Visit https://accounts.hpc.lsu.edu/login_request.php• Enter your INSTITUTIONAL Email Address.• Check your email and click on the link provided (link is active for

24hrs only)• Fill the form provided• For LSU HPC CONTACT/COLLABORATOR field enter the name of your

research advisor/supervisor or course instructor who must be a FullTime Faculty member at LSU. Never enter HPC staff’s name.

• Click Submit button• Your account will be activated once we have verified your credentials


Account Management- LSU HPC and LONI User Portals

Both portals can be found at the top of http://www.hpc.lsu.edu/ LONI account

– https://allocations.loni.org LSU HPC account

– https://accounts.hpc.lsu.edu The default Login shell is bash

– Supported Shells: bash, tcsh, ksh, csh, sh– Change Login Shell at the profile page

May keep the account after graduation under certain circumstances


Allocation

An allocation is a block of service unit (SUs) that allows a user to runjobs on a cluster One SU is one core-hour Example

• 40 SUs will be charged for a job that runs 10 hours on 4 cores LONI & HPC users: All LONI clusters, SuperMikeII and SuperMIC jobs

need to be charged to a valid allocation.


It is FREE to have allocation and use LSU HPC/LONI resources,but there is a value of approximately $0.1/core-hour from othersources (e.g. taxpayer)

Who Can Request Allocation? Only Full Time LSU/LONI Faculty member at LONI member

institutions can act as Principle Investigators (PI) and requestLSU HPC/LONI Allocations. Rule of Thumb: If you can sponsor user accounts, you can

request allocations.


Everyone else will need to join an existing allocation of a PI,usually your advisor or course instructor (if your courserequires a LSU HPC/LONI account).

As a non-PI, your goal is to help your PI understand theallocation policy and prepare the allocation request.

Allocation Types Startup: Allocations upto 50K SUs Can be requested at any time during the year. Begins on the first day of the quarter in which we received the request Only two active allocations per PI at any time.

Large (research): Allocations between 50K - 4M SUs. Decision will be made on January 1, April 1, July 1 and October 1 of each

year A request must be submitted one month before the decision day. Users can have multiple Large Allocations. LSU HPC: Each request is limited to 3 million SUs, and a PI may have a

total of 5 million SUs active at any given time. LONI: Each requests is limited to 4 million SUs, and a PI may have a total

of 6M SUs active at any given time


LSU HPC Research Allocation Proposals Depending on the amount of SUs being requested,

research allocations are categorized into three typesResearch allocation types and requirements


Type Size (SU) Previous usage andoutcome

External funding orLSU demand Number of pages

A 50,000 to300,000 Optional Optional 4

B 300,000 to1,000,000 Required Optional 5

C >1,000,000 Required Required 6

Sample proposals for each type are available now. http://www.hpc.lsu.edu/users/hpcpolicy.php#research LONI allocation procedure is not affected.

Case study – startup allocationUser: “My allocation hpc_xyz was cancelled on 12/31/16, even though theallocation itself was activated less than a year ago (see email below). Couldyou please explain me why?”Forwarded email:From: Accounts Admin [mailto:[email protected]]Sent: Thursday, February 11, 2016 11:11 AMSubject: [CyS-HPC] Project hpc_xyz has been activated


User Services:“The startup allocation begins at the first day of the quarterwhen you applied. So when you applied for this allocation in Feb 2016, itstarts 1/1/16 and ends 12/31/16.”

How to Request/join an Allocation LONI: Login to your profile at https://allocations.loni.org LSU HPC: Login at https://accounts.hpc.lsu.edu/allocations.php Click on "Request Allocation" in the right sidebar Click "New Allocation" to request a New Allocation. (PI only)

Fill out the form provided. All requests require submission of a proposal justifying the use of the resources. Click "Submit Request" button.

Click "Join Allocation" to join an existing Allocation (Non-PI). Search for PI using his/her email address, full name or LONI username Click "Join Projects" button associated with the PI’s information. You will be presented with a list of allocations associated with the PI. Click "Join"

for the allocation you wish to join. Your PI will receive an email requesting him to confirm adding you to the

allocation. Please do not contact the helpdesk to do this.


Case study – startup allocationUser: “Hi, my PI recently applied for an allocation on SuperMike II which wasapproved (see forwarded email below). However, when I view my allocationbalances at https://accounts.hpc.lsu.edu/balances.php, I do not see that thisallocation is available for my use. What steps should I take to access thisallocation?”


User Services:“You should either ask your PI add you to the allocation, orrequest to be added through the user portal”

How Do I Reset My Password?

LONI: Visit https://allocations.loni.org/user_reset.php LSU HPC: Visit https://accounts.hpc.lsu.edu/user_reset.php Enter the email address attached to your account You will receive an email with link to reset your password, link must

be used within 24 hours. Once you have entered your password, your password reset request

needs to be manually reviewed. IMPORTANT:– DO NOT assume your new password is available to use right after the

reset request submission– DO NOT submit the reset request multiple times if you didn’t see your

new password works at once. The Password approval can take anything from 10 mins to a few

hours depending on the schedule of the Admins and also time ofday.

You will receive a confirmation email stating that your passwordreset has been approved.


Case study - password resetUser: “I have been trying to access my accounts on Mike and Queen Bee viaan SSH client, but the connection won't go through. I reset my passwords thisweekend and the terminals keep giving me a "Password Authentication Failed"error message…..”


User Services: “When you send a password reset request, it has to bemanually processed for security reason before your new password becomesavailable.”

Password Security Passwords should be changed as soon as your account is

activated for added security. Password must be at least 12 and at most 32 characters long,

must contain three of the four classes of characters:– lowercase letters,– uppercase letters,– digits, and– other special characters (punctuation, spaces, etc.).

Do not use a word or phrase from a dictionary, Do not use a word that can be obviously tied to the user which

are less likely to be compromised. Do not tell your password to others including your advisor.


During the break… Login to one of the user portals (LSU HPC or LONI) with your HPC

username and password Download MobaXterm (if you are Windows user) Review commands in Linux and the vim editor


Cheat sheet of Commands in Linux


• History• mkdir (name of file) -> makes a folder• ls -> list

-a list all files including hidden-l shows files with a long listing format

• cd -> change directory• pwd -> shows location• cp -> copy• rm -> Remove files (careful)• Up arrow (↑) -> moves back in history• Tab -> fills in unique file name• Tab Tab -> press tab twice, shows all available file names

Cheat sheet of vi


• vi (name of file)• Commands in VI

– i enter insert mode (-- INSERT -- shows in the bottom left corner)– esc exits insert mode, back to the command mode

• dd -> deletes line• u -> Undo• Shift Z shift Z or :wq -> saves and exits VI• :q! -> exit without saving• : (some number) -> moves through file to row #• /(indicator) -> search

– Use N to find Next• [ (page up) ] (page down)

• NO CAPS (for example :q! is not :Q!)

Cluster Nomenclature


Term Definition

Cluster A set of connected computer nodes that work together,each node set to perform the same kind of task (job).

Node A single, named host machine in the cluster.

Core The basic computation unit in the processor (CPU). Forexample, a quad-core processor has 4 cores.

Job A user's request to use a certain amount of resources for acertain amount of time on cluster for his/her work.

40

General Cluster Architecture Multiple compute nodes Multiple users Each user may have multiple jobs running simultaneously


Accessing Cluster via SSH (Secure Shell) On Linux and Mac

– use ssh on a terminal to connect


Windows box (ssh client):– MobaXterm (recommended)– SSH Secure Shell Client– Putty (Better use Bitvise SSH Client from http://www.putty.org/)

Username and password Host name

– LONI: <cluster_name>.loni.org• <cluster_name> can be:

– qb.loni.org

– LSU HPC: <cluster_name>.hpc.lsu.edu• <cluster_name> can be:

– mike.hpc.lsu.edu– smic.hpc.lsu.edu– philip.hpc.lsu.edu

Accessing Cluster on Linux and Mac


Accessing Cluster on Windows - MobaXterm


First time user, choose either one:– use ssh on a terminal– start a new remote session -> SSH

Choose Noif you areunsure

Accessing cluster on Windows - SSHSecure Shell Client

SSH Secure Shell Client– command line scp and rsync– sftp file transfer through GUI


Accessing cluster on Windows - Putty


Cluster Environment Multiple compute nodes Multiple users Each user may have multiple jobs running simultaneously


DO NOT RUNYOUR JOB ONTHE HEADNODE!!!

Enable X11 Forwarding On Linux or Mac, simply pass the -X option to the ssh command line

– ssh -X [email protected] On Windows using putty

– Connection->SSH->X11->Enable X11 forwarding– Install X server (e.g. Xming)


On Windows using MobaXterm– X server already set up– Automatically start X server at

start up (Settings -> X11)

Cluster Environment Useful commands on the head node

– pwd, print current directory– who, check who is on the node– balance, check allocation balance


File Systems


Tips– /home/$USER, /work/$USER, /project/$USER– The work directory is not for long-term storage

• Files are subject to be purged after 60-90 days– The work directory will be created 1 hour after the first cluster login– The project is not available by default

• Only PI can apply storage allocation to use and renew periodically• NOT for archival purposes• For more info about the storage allocation, see appendix or contact us

Directory(folder) Distributed Throughput File life

time Best used for

Home Yes Low Unlimited Code in development, compiledexecutable

Work Yes High 60 days Job input/output

Project Yes Medium/High 1 yearstorage space for a specificproject, NOT for archival

purposes

50

Disk Quota

The work directory will be created within an hour after the first login– “To stress the importance of this point, I will repeat it again”

Never let you job write output to your home directory Check current disk quota and usage

– showquota


ClusterHome Work Local scratch

Access point Quota Access Point Quota Access point

LONI/home/$USER 5 GB /work/$USER N/A /var/scratch

HPC

51

File Transfer (Linux/Mac) From/to a Unix/Linux/Mac machine (including between the clusters)

– scp command• Syntax: scp <options> <source> <destination>

– rsync command• Syntax: rsync <options> <source> <destination>

[ychen64@mike2 ~]$ scpusage: scp [-1246BCpqrv] [-c cipher] [-F ssh_config] [-i identity_file]

[-l limit] [-o ssh_option] [-P port] [-S program][[user@]host1:]file1 ... [[user@]host2:]file2

[ychen64@mike2 ~]$ rsyncrsync version 3.0.6 protocol version 30Usage: rsync [OPTION]... SRC [SRC]... DEST

From a download link on a website (usually opened with a web browser)– Right click on the link and then copy the link location– wget command

[ychen64@mike2 ~]$ wget <paste_your_copied_link_here>


File Transfer (Windows) From/to a Windows machine

– Use a client that supports the scp protocol (e.g. SSH Secure Shell Client)


Review the use of vim editor


• vi (name of file)• Commands in VI

– i is insert mode (-- INSERT -- shows in the bottom left corner)– esc exits insert mode, back to the command mode

• dd -> deletes line• u -> Undo• Shift Z shift Z or :wq -> saves and exits VI• :q! -> exit without saving• : (some number) -> moves through file to row #• /(indicator) -> search

– Use N to find Next• [ (page up) ] (page down)

• NO CAPS (e.g. :q! ≠ :Q!)

Application Software Installed Software

– Mathematical and utility libraries• FFTW, HDF5, NetCDF, PETSc...

– Applications• Amber, CPMD, NWChem, NAMD, Gromacs, R, LAMMPS...

– Visualization• VisIt, VMD, GaussView

– Programming Tools• Totalview, DDT, TAU...

List of software– http://www.hpc.lsu.edu/resources/software/index.php

Installed under /usr/local/packages User requested packages

– Usually installed in user home directory, unless request by a group ofusers, in which case it will be installed under /project or/usr/local/packages


Software Environment: Module and Softenv Environment variables

– PATH: where to look for executables– LD_LIBRARY_PATH: where to look for shared libraries– LD_INCLUDE_PATH: where to look for header and include files

Other environment variables sometimes needed by various software– LIBRARY_PATH, C_LIBRARY_PATH– LDFLAGS, LDLIBS

SoftEnv– A software that helps users set up environment variables properly to

use other software package. Much more convenient than settingvariables in .bashrc

– SuperMike2 Modules

– Another software that helps users set up their environment. Mostsupercomputing sites (including XSEDE) use modules.

– SuperMIC, Philip and QB2


Softenv: Listing All Packages Command “softenv” lists all packages that are managed by SOFTENV


softenv key

softenv on SuperMike IIexample

57

Softenv: Searching A Specific Package Use “-k” option with softenv”:

– softenv -k fftw

Or use grep with softenv– softenv | grep “fftw”


Softenv: One time change of environment Set up the environment variables to use a certain software package in

the current login session only– Add a package: soft add <key>– Remove a package: soft delete <key>


Softenv: Permanent change of environment Set up the environment variables to use a certain software package

– First add the key to ~/.soft– Then execute resoft at the command line– The environment will be the same next time you log in


Exercise: Use Softenv Find the key for gromacs-4.5.5

– Set up your environment to use gromacs-4.5.5 (one time change)– Check if the variables are correctly set by “which mdrun”

Find the key for Python-2.7.3– Set up your environment to permanently use Python-2.7.3– Check if the variables are correctly set by “which python”


Using Environment Modules Similar to Softenv, Environment Modules is another framework to

manage what software is loaded into a user's environment. Itsfunctionality includes– List all software packages currently available in the Environment

Modules system,– List all software packages loaded into a user's environment,– Load/Switch software packages into a user's environment– Unload a software package from a user's environment.


Modules: List All Available Packages The command to list all available packages is: module avail/av[fchen14@smic1 ~]$ module av------------- /usr/local/packages/Modules/modulefiles/apps -------------------abyss/1.5.2/INTEL-140-MVAPICH2-2.0 mpich/3.1.1/INTEL-14.0.2ansys/15.0 mummer/3.23/INTEL-14.0.2beast/1.7.5 mumps/4.10.0/INTEL-140-MVAPICH2-2.0blast/2.2.22 muscle/3.8.31blast/2.2.28/INTEL-14.0.2 mvapich2/2.0/INTEL-14.0.2(default)...---------------- /usr/local/packages/Modules/modulefiles/xsede -----------------ant/1.9.4 java/1.7.0 tgresid/2.3.4 unicore/6.6.0globus/5.0.4-r1 pacman/3.29-r3 tgusage/3.0 xsede/1.0(default)gx-map/0.5.3.3-r1 tginfo/1.1.4 uberftp/2.6

The format of the listed packages is <package name>/<packageversion>. For example, gcc/4.4.2 is version 4.4.2 of gcc.


Modules: List Currently Loaded Packages To see what packages are currently loaded into a user's environment,

the command is: module list

[fchen14@smic1 ~]$ module listCurrently Loaded Modulefiles:

1) intel/14.0.2 6) tginfo/1.1.4 11) xsede/1.02) mvapich2/2.0/INTEL-14.0.2 7) ant/1.9.4 12) ansys/15.03) gx-map/0.5.3.3-r1 8) java/1.7.0 13) EasyBuild/1.13.04) tgusage/3.0 9) uberftp/2.6 14) INTEL/14.0.25) globus/5.0.4-r1 10) tgresid/2.3.4 15) impi/4.1.3.048/intel64

The above listing shows that this user has 15 packages loaded


Modules: Load/Unload a Package The command for loading a package into a user's environment is:

module load <package name>. The command for unloading a package is: module unload <package

name>. If a specific version of a package is desired, the command can be

expanded to: module load <package name>/<package version>.

[fchen14@smic1 ~]$ module av intel-------------- /usr/local/packages/Modules/modulefiles/apps ---------------------intel/13.0.0 intel/13.1.3 intel/14.0.2(default)[fchen14@smic1 ~]$ module load intel[fchen14@smic1 ~]$ icc -vicc version 14.0.2 (gcc version 4.4.7 compatibility)[fchen14@smic1 ~]$ module unload intel[fchen14@smic1 ~]$ module load intel/13.1.3[fchen14@smic1 ~]$ icc -vicc version 13.1.3 (gcc version 4.4.7 compatibility)


Modules: Unload All Loaded Packages To unload all loaded module files, use the purge method:

[fchen14@smic1 ~]$ module listCurrently Loaded Modulefiles:

1) intel/14.0.2 3) r/3.1.0/INTEL-14.0.2 5) ansys/15.02) INTEL/14.0.2 4) impi/4.1.3.048/intel64

[fchen14@smic1 ~]$ module purge[fchen14@smic1 ~]$ module listNo Modulefiles Currently Loaded.[fchen14@smic1 ~]$


Modules: Dependencies Note that Modules will load any prerequisites (dependencies) for a

package when that package is loaded. (SoftEnv does not)

[fchen14@smic1 ~]$ module listNo Modulefiles Currently Loaded.[fchen14@smic1 ~]$ module av gromacs

---------------- /usr/local/packages/Modules/modulefiles/apps -------------------gromacs/5.0/INTEL-140-MVAPICH2-2.0(default)[fchen14@smic1 ~]$ module load gromacs/5.0gromacs/5.0 gromacs/5.0/INTEL-140-MVAPICH2-2.0[fchen14@smic1 ~]$ module load gromacs[fchen14@smic1 ~]$ module listCurrently Loaded Modulefiles:

1) intel/14.0.2 3) mvapich2/2.0/INTEL-14.0.22) INTEL/14.0.2 4) gromacs/5.0/INTEL-140-MVAPICH2-2.0


Modules: Display the module changes The display/show command will detail all changes that will be made to

the user’s environment: module disp <package name>.[fchen14@smic1 ~]$ module disp python/2.7.7-anaconda-------------------------------------------------------------------/usr/local/packages/Modules/modulefiles/apps/python/2.7.7-anaconda:module-whatis Description: Python is a programming language that lets you workmore quickly and integrate your systems more effectively. - Homepage:http://python.org/conflict pythonprepend-path CPATH /usr/local/packages/python/2.7.7-anaconda/includeprepend-path LD_LIBRARY_PATH /usr/local/packages/python/2.7.7-anaconda/libprepend-path LIBRARY_PATH /usr/local/packages/python/2.7.7-anaconda/libprepend-path MANPATH /usr/local/packages/python/2.7.7-anaconda/share/manprepend-path PATH /usr/local/packages/python/2.7.7-anaconda/binprepend-path PKG_CONFIG_PATH /usr/local/packages/python/2.7.7-anaconda/lib/pkgconfigprepend-path PYTHONPATH /usr/local/packages/python/2.7.7-anaconda/lib/python2.7/site-packagessetenv LHPC_ROOTPYTHON /usr/local/packages/python/2.7.7-anacondasetenv LHPC_VERSIONPYTHON 2.7.7-------------------------------------------------------------------


Modules: Load Automatically on Login On HPC and LONI clusters, Modules can be loaded automatically on

login by adding the appropriate module load commands to a user's~/.bashrc or ~/.modules (recommended) file

The following example shows a .modules file that automatically loadsR, intel mpi and ansys-15.0[fchen14@smic1 ~]$ cat ~/.modules## This is the default .modules file for smic# It is used to customize your Modules environment# variables such as PATH and LD_LIBRARY_PATH.## The default software stack on smic uses# Intel/cluster_studio_xe_2013.1.046/composer_xe_2013_sp1.2.144# mvapich2/2.0/INTEL-14.0.2## To learn more about available software, try:# module --help## Your default softwaremodule load r/3.1.0/INTEL-14.0.2module load impi/4.1.3.048/intel64module load ansys/15.0# You can add additional software here


Creating Your Own Module File An example of a simple module file (~/my_module/gitkey):#%Moduleproc ModulesHelp { } {

puts stderr { my compiled version of git.}

}module-whatis {version control using git}set GIT_HOME /home/fchen14/packages/git-master/installprepend-path PATH $GIT_HOME/bin Add the path to the key to the MODULEPATH environment variable:$ export MODULEPATH=~/my_module:$MODULEPATH

Then try to use:$ module load gitkey$ which git$ module unload gitkey$ which git


Exercise: Use Modules Find the key for gromacs-4.5.5

– Set up your environment to use gromacs-4.5.5 (one time change)– Check if the variables are correctly set by “which mdrun”

Find the key for Python-2.7.3– Set up your environment to permanently use Python-2.7.3– Check if the variables are correctly set by “which python”


Two Job Types Interactive job

– Set up an interactive environment on compute nodes for users• Advantage: can run programs interactively• Disadvantage: must be present when the job starts

– Purpose: testing and debugging• DO NOT RUN YOUR JOB ON THE HEAD NODE!!!• Try not to run interactive jobs with large core count, which is a waste of

resources) Batch job

– Executed without user intervention using a job script• Advantage: the system takes care of everything• Disadvantage: can only execute one sequence of commands which cannot

changed after submission– Purpose: production run


Submitting Jobs on Linux Clusters Interactive job example:

qsub -I -V \-l walltime=<hh:mm:ss>,nodes=<num_nodes>:ppn=<num_cores> \-A <Allocation> \-q <queue name>

Note: ppn=16 for SuperMike2; ppn=20 for QueenBee2 and SuperMic Add -X to enable X11 forwarding


Job Monitoring - Linux Clusters Check details on your job using qstat

$ qstat -f jobid : For details on your job$ qstat -n -u $USER : For quick look at nodes assigned to you$ qdel jobid : To delete job

Check approximate start time using showstart$ showstart jobid

Check details of your job using checkjob$ checkjob jobid

Check health of your job using qshow$ qshow -j jobid

Pay close attention to the load and the memory consumed by yourjob!


Exercise Run an interactive job session for 30 min, using nodes=1:ppn=16

– Verify using hostname that you are not on the head node– Download the tarball from HPC website to the home directory– $ tar –xvzf a.tar.gz :untar it– $ cd serialpi : cd to the directory “serialpi”– $ cp a.out /work/$USER :Copy it to the work directory– $ cd /work/$USER : cd to your work directory– $ ./a.out ##### :Execute a.out along with a number argument

At job juntime– Open a new terminal and login to the head node– $ qstat -u $USER : Find out your current job id and status– $ qstat -n jobid : Quick look at the compute node assigned to you– $ qshow -j jobid : Check health of your job using qshow– $ ssh mike/qb/smic### :SSH to (your own) compute node– Now on the compute node, check– $ free –m Check free memory– $ top Monitoring job


Job management Job management basics

– Find appropriate queue– Understand the queuing system and your requirements and proceed to

submit jobs– Monitor jobs


Job Queues Nodes are organized into queues. Nodes can be shared. Each job queue differs in

– Number of available nodes– Max run time– Max running jobs per user– Nodes may have special characteristics: GPU’s, Large memory, etc.

Jobs need to specify resource requirements– Nodes, time, queue

Its called a queue for a reason, but jobs don’t run on a “First ComeFirst Served” policy


Queue Characteristics – LONI clusters


Machine Queue MaxRuntime ppn

Maxrunning

jobs

Maxnodesper job

Use

Eric

workq

3 days

816

24 Unpreemptable

checkpt 8 48 Preemptable

single 1 32 1 ppn < =8

QB2

workq3 days

20

44

128 Unpreemptable


single 7 days 1,2,4,8 1 Single nodejobs

78

Queue Characteristics – LSU Linux clusters


Machine Queue MaxRuntime ppn

Maxrunning

jobs

Maxnodes per

jobUse

SuperMike II

workq3 days

16

34

128 Unpreemptable


bigmem 2 days 16 1 Big memory

gpu 3 days 16 16 Job using GPU

single 3 days 1,2,4,8 1 Single nodejobs

Philip

workq

3 days

8

5

4 Unpreemptable


bigmem 8 2 Big memory

single 14 days 4 50 1 Singleprocessor

SuperMICworkq

3 days20

34128 Unpreemptable


79

Queue Characteristics “qstat -q” will give you more info on the queues[fchen14@mike2 ~]$ qstat -q

server: mike3

Queue Memory CPU Time Walltime Node Run Que Lm State---------------- ------ -------- -------- ---- --- --- -- -----workq -- -- 72:00:00 128 31 6 -- E Rmwfa -- -- 72:00:00 8 3 0 -- E Rbigmem -- -- 48:00:00 1 0 0 -- E Rlasigma -- -- 72:00:00 28 28 7 -- E Rbigmemtb -- -- 48:00:00 1 0 0 -- E Rpriority -- -- 168:00:0 128 0 0 -- E Rsingle -- -- 72:00:00 1 62 0 -- E Rgpu -- -- 24:00:00 16 1 0 -- E Rpreempt -- -- 72:00:00 -- 0 0 -- E Rcheckpt -- -- 72:00:00 128 31 137 -- E Radmin -- -- 24:00:00 -- 0 0 -- E Rscalemp -- -- 24:00:00 1 0 0 -- E R

----- -----156 150

For a more detailed description use mdiag


Queue Querying – Linux Clusters Displays information about active, eligible, blocked, and/or recently

completed jobs: showq command[fchen14@mike2 ~]$ showqactive jobs------------------------JOBID USERNAME STATE PROCS REMAINING STARTTIME294690 michal Running 64 00:22:41 Sat Jan 24 23:05:26294693 michal Running 64 00:22:41 Sat Jan 24 23:05:26...183 active jobs 6399 of 7512 processors in use by local jobs (85.18%)

401 of 468 nodes active (85.68%)eligible jobs----------------------JOBID USERNAME STATE PROCS WCLIMIT QUEUETIME

295775 stumusii Idle 32 3:00:00:00 Tue Jan 27 21:15:35295776 stumusii Idle 32 3:00:00:00 Tue Jan 27 21:17:43...9 eligible jobsblocked jobs-----------------------JOBID USERNAME STATE PROCS WCLIMIT QUEUETIME0 blocked jobsTotal jobs: 192

Command: qfree, qfreeloni– Show the number of free, busy and queued


Take-home message All words in red font are very important, CAPITALIZED ARE EVEN

MORE IMPORTANT! What is HPC and how to evaluate the computational speed (FLOPS) Available HPC resources

– LSU HPC, LONI, XSEDE– SuperMike2, SuperMic, Queenbee, Philip, Eric

Account eligibility, allocation, password reset procedure.– Need account sponsor (most likely a faculty) to support HPC account– It is free to run your simulation on cluster, but you need allocation

Infrastructure– What is (1) node, (2) core (3) cluster (4) job– Where are you upon login to the cluster, where your job should be on

Practice on the cluster– How to login via SSH– How to check your quota and balance– How to transfer files, add software by Softenv or Modules


Next Week Training HPC User Environment 2, February 21

– More on job management and queue– Compiling serial/parallel program

Weekly trainings during regular semester– Wednesdays “9:00am-11:00am” session, Frey 307 CSC

Programming/Parallel Programming workshops– Usually in summer

Keep an eye on our webpage: www.hpc.lsu.edu


HPC@LSU User Services• Hardware resources

– Currently manages 4 clusters• Software stack

– Communication software– Programming support: compilers and libraries– Application software

• Contact user services• Email Help Ticket: [email protected]• Telephone Help Desk: +1 (225) 578-0900


AppendixStorage Allocation on /project

Don’t be confused with the computational allocation introduced in thistraining

PI can apply for extra disk space on the /project volume for you andhis/her entire research group if– your research requires some files to remain on the cluster for a fairly

long period of time; and– their size exceeds the quota of the /home

The unit is 100 GB Storage allocations are good for 6 months, but can be extended based

on the merit of the request Examples of valid requests

– I am doing a 6-month data mining project on a large data set– The package I am running requires 10 GB of disk space to install

Examples of invalid requests– I do not have time to transfer the data from my scratch space to my

local storage and I need a temporary staging area


AppendixComputing an approximate value for PI

The a.out in this training calculates the value for PI based on the math which isactually quite simple: Imagine a square dartboard with circle inscribed within it suchthat the diameter of the circle is the length of a side of the square.

We can observe that the ratio of the area of the circle to the area of the square isequal to some constant, π/4 (since the square’s area is 2*2 = 4 and area_circle =π*r^2 = π). If we randomly place many points (darts) inside the square, we can counthow many are also inside the circle (satisfy x^2+y^2 <= 1) vs the total number ofpoints and compute an estimate for the value of π. (Problem description is fromJared Baker, UW; Ben Matthews, NCAR)


hpc user environment 1...hpc user environment 1 yuwu chen hpc user services lsu hpc loni...

Documents